Professional Documents
Culture Documents
2008
COMPACT MODELING OF DEEP SUBMICRON CMOS TRANSISTOR WITH
SHALLOW TRENCH ISOLATION MECHANICAL STRESS EFFECT
by
August 2008
ACKNOWLEDGEMENTS
I would like to express my gratitude to my Ph.D. project supervisor, Dr. Othman Sidek
for his guidance and encouragement throughout the duration of this project. His
support in determining the project that benefit to both my work and study is highly
appreciated.
encouragement for further study and technical publications. Not forgetting to all my
for everything.
beloved family, especially to my wife, Christine. For their undying love, support and
ii
TABLE OF CONTENTS
Page
ACKNOWLEDGEMENTS ii
TABLE OF CONTENTS iii
LIST OF TABLES vii
LIST OF FIGURES viii
LIST OF SYMBOL xiv
LIST OF ABBREVIATION xx
ABSTRAK xxii
ABSTRACT xxiv
CHAPTER 1 : INTRODUCTION
iii
1.3.17 Effective Width and Effective Length 30
1.4 Why Focus on STI Mechanical Stress Effect Modeling? 30
1.5 Background and Objective 31
1.6 Contributions of the Proposed Compact Models 32
1.7 Chapter Summary 33
iv
CHAPTER 4 : MODELING OF STI MECHANICAL Y-STRESS
EFFECT
REFERENCES 117
PUBLICATIONS 123
v
APPENDICES
A Compact STI x-Stress HSPICE Model 126
for 0.13 μm Technology 1.2 V CMOS Transistor
B Empirical-Based STI x-Stress Width Dependence HSPICE 129
Model for 0.13 μm Technology 1.2 V CMOS Transistor
C Empirical-Based STI x-Stress Mismatch HSPICE Model 132
for 0.18 μm Technology 1.8 V CMOS Transistor
D Physical-Based STI y-Stress HSPICE Model 135
for 0.13 μm Technology 1.2 V High-Vt CMOS Transistor
E Physical-Based STI y-Stress HSPICE Model 138
for 0.13 μm Technology 2.5 V CMOS Transistor
vi
LIST OF TABLES
Page
Table 3.1 Summary of short channel Vtlin change due to STI x-stress 53
for 0.18 μm CMOS technology. Published in (Tan et al.,
2004a and Tan et al., 2004b).
Table 3.2 Summary of short channel Idsat change due to STI x-stress 55
for 0.18 μm CMOS technology. Published in (Tan et al.,
2004a and Tan et al., 2004b).
Table 3.4 Summary of short channel Vtlin change due to STI x-stress 61
for 0.13 μm CMOS technology.
Table 3.5 Summary of short channel Idsat change due to STI x-stress 63
for 0.13 μm CMOS technology.
vii
LIST OF FIGURES
Page
Figure 1.2 Charge regions: (a) in long channel; (b) in short channel 8
(Liu, 2001).
Figure 1.7 Surface potential versus L for (a) long channel transistor 14
(b) short channel transistor, and (c) short channel
transistor at high Vd (Taur and Ning, 1998).
Figure 1.9 CMOS transistor operates in; (a) linear region, (b) onset 17
of saturation, (c) beyond saturation where CLM occurs
(Taur and Ning, 1998).
Figure 1.10 Id-Vd curves with Vg,2 > Vg,1, showing the relationship 18
between Va and the slope of Id in saturation (Liu, 2001).
Figure 1.15 SCBE effect that increases Isub and Id due to impact 25
ionization (Cheng and Hu, 1999).
viii
Figure 1.16 Id-Vd and Rout characteristics to illustrate how their 26
behaviors are dominated by CLM, DIBL and SCBE
(Cheng and Hu, 1999).
Figure 1.17 NMOS drain region when the channel is (a) inverted, (b) 27
accumulated and (c) when n+ region surface is
depleted or inverted (Taur and Ning, 1998).
Figure 1.18 Poly gate depletion region (Cheng and Hu, 1999). 28
Figure 2.2 NMOS Idsat versus layout types. Idsat decrease in small 35
type L = 0.2 μm is equivalent to 30 nm increase in L for
medium type L = 0.2 μm (Scott et al., 1999).
Figure 2.7 Idsat change due to external applied stress for (a) PMOS 38
and (b) NMOS with 500 μm gate length (Chen and
Huang, 2002).
Figure 2.11 Distance between STI edge and gate edge, “a” (Bianchi 43
et al., 2002).
ix
Figure 3.1 Compressive STI mechanical stress on a CMOS 51
transistor.
Figure 3.4 Effect of STI y-stress on wide width and narrow width 56
transistors.
Figure 3.8 Short channel Vtlin and Idsat versus sa characteristics for 62
0.13 μm CMOS technology. N = 27 dice. Published in
(Tan et al., 2005a and Tan et al., 2006a).
x
Figure 3.16 Possible mechanism of STI x-stress effect on CMOS 73
transistors mismatch.
Figure 4.1 Actual silicon data showing a hook shaped Idsat curve. 78
Figure 4.5 PMOS Idsat versus width curve with and without STI y- 83
stress. N = 6 dice.
Figure 4.6 NMOS Idsat versus width curve with and without STI y- 83
stress. N = 6 dice.
Figure 4.14 Effect of the STI y-stress parameters in fitting the hook 93
shaped Idsat curve.
xi
Figure 4.15 Parameters that control different regions of the hook 94
shaped Idsat curve. Symbol is data and line is model.
Figure 4.21 Fitting of the proposed STI y-stress model on 2.5 V 100
NMOS Vtlin and Idsat. N = 1 dice.
Figure 4.22 Fitting of the proposed STI y-stress model on 2.5 V 101
PMOS Vtlin and Idsat. N = 1 dice.
Figure 5.9 Conduction band splitting due to compressive x-, y- and 108
z-stress.
xii
Figure 5.12 Holes redistribution in the constant energy valleys of 110
valence band due to compressive x-, y- and z-stress.
Figure 5.13 Constant energy valleys of conduction band for narrow 111
width transistor under STI compressive x-stress.
Figure 5.14 Constant energy valleys of valence band for narrow 112
width transistor under STI compressive x-stress.
xiii
LIST OF SYMBOLS
% Percent
≥ Greater or equal
Δ Delta
π pi (3.142)
μ Carrier mobility
η Ideality factor
μA micro-Ampere
μm micrometer
δNP Positive sign for NMOS and negative sign for PMOS
< Smaller
a Active space
C Capacitance
d Distance
xiv
E Electric Field
I Current
Id Drain current
J Current density
k p x (h/2π)
L Channel length
m meter
m* Effective mass
mA milli-Ampere
meV milli-electron-Volt
mV milli-Volt
N Number of samples
nc Carrier concentration
nm nanometer
p Momentum
xv
Pags Gate bias dependence bulk charge parameter
xvi
Pketa Body bias dependence bulk charge parameter
Pll, Plw, Plwl, Plln, Plwn Channel length geometry scaling parameters
xvii
Pstxvth01 STI x-stress Vt parameter
Pwl, Pww, Pwwl, Pwln, Pwwn Channel width geometry scaling parameters
Rd Drain resistance
xviii
Rds Drain and source resistance
Rs Source resistance
sa, sb Distance between STI edge and gate edge (active space)
T Temperature
V Voltage
Va Early voltage
Vb Bulk voltage
Vd Drain voltage
Vg Gate voltage
Vs Source voltage
Vt Threshold voltage
W Channel width
xix
LIST OF ABBREVIATION
2D Two-dimensional
3D Three-dimensional
AA Active Area
CD Critical Dimension
DC Direct Current
DW Delta Width
exp Exponential
xx
IC Integrated Circuit
SD Source/drain
Si Silicon
SS Small Size
xxi
PEMODELAN PADAT UNTUK TRANSISTOR CMOS DI BAWAH
SUBMIKRON DENGAN KESAN TEKANAN MEKANIKAL PENGASINGAN
PEPARIT CETEK
ABSTRAK
Thesis ini memperkenalkan satu model padat, dua model berasaskan empirikal
dan satu model berasaskan fizikal untuk kesan tekanan mekanikal Pengasingan
Peparit Cetek (STI) ke atas transistor CMOS di bawah submikron. Model tekanan-x
STI padat digunakan untuk menangkap tekanan dalam arah laluan transistor panjang.
Model ini adalah lebih ringkas berbanding dengan model tekanan STI BSIM4, tetapi
dapat mencapai ketepatan yang serupa. Dua ciri-ciri baru tekanan-x STI telah
dikenalpastikan. Ciri yang pertama adalah fakta bahawa kesan tekanan-x STI bagi
transistor CMOS berubah untuk transistor lebar yang berlainan. Kesan ini telah dikesan
dalam teknologi-teknologi 0.18 μm dan 0.13 μm. Satu model kesan tekanan-x STI
bergantung kepada lebar yang empirikal telah dicadangkan untuk menangkap kesan
ini. Ciri baru yang kedua adalah fakta bahawa kesan tekanan-x STI mengubah ciri-ciri
ketidaksamaan transistor CMOS. Satu model Monte Carlo yang empirikal telah
dicadangkan untuk menangkap kesan ini. Satu lengkuk arus parit tepu (Idsat)
berupabentuk cangkul berlawan dengan transistor lebar yang baru telah dikenalpasti.
Lengkuk ini tidak dapat dimodelkan dengan mengunakan model tekanan STI BSIM4.
Dengan mengunakan satu kaedah bentangan yang baru, ciri-ciri fizikal lengkuk ini
telah dikenalpasti. Lengkuk Idsat berupabentuk cangkul ini disebabkan oleh gabungan
kesan-kesan daripada tekanan-y STI (tekanan STI dalam arah laluan transistor lebar)
yang menurunkan Idsat dan kesan Perubahan Lebar (DW) yang meningkatkan Idsat.
Berdasarkan ciri-ciri fizikal ini, satu model tekanan-y STI yang fizikal diperkenalkan
yang dicadangkan di dalam thesis ini dipastikan dengan mengunakan silikon data
μm dan 0.13 μm yang berpiawaian industri. Model-model baru ini dibina dalam model
xxii
BSIM3v3 dengan mengunakan kaedah model makro (juga dikenali sebagai kaedah
sublitar). Dua parameter SPICE, parameter voltan ambang tiada pincang belakang,
Pvth0 dan parameter kelincahan pembawa, Pu0, telah digunakan untuk membina model-
model ini. Perbezaan masa simulasi antara model makro dan model konventional
xxiii
COMPACT MODELING OF DEEP SUBMICRON CMOS TRANSISTOR
WITH SHALLOW TRENCH ISOLATION MECHANICAL STRESS EFFECT
ABSTRACT
deep submicron CMOS transistor. The compact STI x-stress model is used to capture
the stress effect in the channel length direction. This model is simpler than the BSIM4
STI stress model, but able to achieve the similar accuracy. Two new characteristics of
STI x-stress have been identified. The first characteristic is the fact that the STI x-
stress effect on CMOS transistor varies for different transistor channel widths. An
empirical width dependence of STI x-stress effect model has been proposed to capture
this effect. The second new characteristic is the fact that STI x-stress effect changes
proposed to capture this effect. A new hook shaped saturation drain current, Idsat curve
versus channel width has been identified. This curve cannot be modeled using the
BSIM4 STI stress model. By using a new layout method, the physical characteristics of
the curve are identified. The hook shaped Idsat curve is caused by the combined effects
of STI y-stress (stress in the channel width direction) that degrades the Idsat and the
Delta Width (DW) effect that increases the Idsat. Based on the physical characteristics, a
new physical-based STI y-stress model is proposed to capture the hook shaped Idsat
behavior. The accuracy of the models in this thesis is verified on actual silicon data
These new models are incorporated into the BSIM3v3 model by using macro model
method (also known as subcircuit method). The two SPICE parameters, the zero back
bias threshold voltage parameter, Pvth0 and the carrier mobility parameter, Pu0, are used
for developing these models. The difference in simulation time between the macro
xxiv
CHAPTER 1
INTRODUCTION
This chapter introduces the compact modeling works in this thesis. First, the
modeling of CMOS transistor which started with the theory Gradual Channel
Approximation (GCA) is reviewed. How the important deep submicron effects of CMOS
After reviewing the theoretical concept of compact modeling, the reasons why
this research focuses on SPICE modeling of the STI stress effect on CMOS transistor
is discussed. The background and objective of this research are then given. Finally, the
research is discussed.
the behaviors of semiconductor devices and able to run with SPICE simulator at
minimum time. Compact modeling is finding the parameter values for the compact
model equations to achieve a good fit between the simulation data and the measured
data. The main objective of the compact modeling is to enable the accurate prediction
When transistor scaling goes to deep submicron, more and more significant
that are negligible in the CMOS micron technology have dominated the behavior of the
Many researchers have put in much effort in establishing the set of equations
(Compact Model) that can provide the most accurate circuit simulation at the minimum
1
time. Berkeley Short-Channel IGFET models (BSIM3v3 and BSIM4 models) are
currently widely used by wafer fabrication foundries and IC design houses. Other
SPICE models such as PSP model, EKV model, HISIM model and XSIM model are
All the modeling works in this thesis is based on BSIM3v3.24, where the final
released version is BSIM3v3.3 model (Cheng et al., 2005). In short, these BSIM3v3
family models, which built on similar source code, are known as BSIM3v3 model in this
thesis. The discussion in this thesis also references to BSIM4.3.0 (Xi et al., 2003) and
BSIM4.6.1 (Dunga et al., 2007) models for STI stress model discussion. In short, all the
BSIM4 family models are known as BSIM4 models in this thesis. The BSIM4 model is
built on a different source code compared to the BSIM3v3. Hence, the BSIM4 model is
not backward compatible to the BSIM3v3 model. The technology nodes of the
experimental transistors that are used in this thesis are 0.18 μm and 0.13 μm.
transistors, how much current is carried in the “on” state of a transistor and how much
leakage current flows in the “off” state must be calculated. In short, the current-voltage,
I-V and capacitance-voltage, C-V characteristics of the CMOS transistor are needed,
specifically the quantitative relationship between the drain current and terminal
voltages.
In the beginning of the compact modeling era, there were four long-channel
explained in the reference book (Wolf, 1995). All the four models are based on the
long-channel Gradual Channel Approximation (GCA), which assumes that the variation
of the electric field in the x-direction (channel length, L direction) is much less than the
2
corresponding variation in the z-direction (perpendicular to channel plane direction).
The four models are Bulk-Charge model, Square-Law model, Pao-Sah model and
Charge-Sheet model.
Bulk-Charge model (Ihantola and Moll, 1964) and Square-Law model (simplified
version of Bulk-Charge model) are only valid if the CMOS transistor is operated in
Pao-Sah model (Pao and Sah, 1966) is the first model that able to cover the
entire range of CMOS transistor operation. This model retains the GCA but the drain
complexity has limited the role of this model for theoretical analysis of the CMOS
transistor.
The limited practical usage of the Pao-Sah model motivated a search for an
approximate advanced analytical model that is still accurate over a wider range of
operating conditions. Charge-Sheet model (Brews, 1978) has become the most widely
computational effort than that of the Pao-Sah model, the computational effort for
determining Id is still too great for most circuit simulation applications, and thus the
Charge-Sheet model has not been incorporated into the popular circuit simulation
region, the simpler Bulk-Charge and Square-Law models are used. The combination of
these models enables the SPICE simulator to cover the entire range of CMOS
When comparing the usage of the Bulk-Charge and Square-Law models, the
primary advantage of the Square-Law model is its simplicity. Most hand calculations for
3
circuit design with CMOS transistors make use of the Square-Law model. It is also the
model used for the simplest (Level 1) analysis of circuits in SPICE simulator. While the
more-accurate Level 2 SPICE for CMOS transistor is based on the Bulk-Charge model.
This section discusses how SPICE model equations able to capture the CMOS
transistors deep submicron effects. BSIM3v3 modeling equations are selected for the
discussion in this section because the modeling equations are less complicated
compared to BSIM4 but sufficient to describe most of the CMOS transistor deep
submicron effects. One can easily catch up with BSIM4 or others SPICE models with
drain current, Id, off current, Ioff and substrate current, Isub. How the deep submicron
effects change the transistor electrical characteristics and how these effects can be
are based on three SPICE modeling reference books (Taur and Ning, 1998, Cheng and
Hu, 1999, and Liu, 2001), and two BSIM manuals; BSIM3v3 manual (Cheng et al.,
section, are (1) Body effect, (2) Short Channel Vt Effect, SCE due to charge sharing,
(3) Reverse Short Channel Effect, RSCE due to lateral non-uniform doping, (4) Narrow
Width Effect, NWE, due to LOCOS isolation, (5) Inverse Narrow Width Effect, INWE
due to STI isolation, (6) Small Size effect, SS, (7) Drain Induced Barrier Lowering,
DIBL, (8) Channel Length Modulation, CLM, (9) Velocity Saturation, (10) Subthreshold
conduction, (11) Field dependent mobility, (12) Substrate Current induced Body Effect,
SCBE, (13) Gate Induced Drain Leakage, GIDL, (14) Polysilicon gate depletion effect,
(15) Velocity overshoot, (16) Source and drain resistance effect and (17) Effective
4
In BSIM3v3, there are six effects that change the threshold voltage, Vt as
shown in Equation 1.
where Pvth0 is the BSIM3v3 threshold voltage parameter of a long channel transistor at
zero back bias, Vb = 0 V. δnp is positive sign for NMOS and negative sign for PMOS.
The four effects that increase Vt are body effect, RSCE, NWE and SS. The other two
effects that decrease Vt are SCE and DIBL. These six effects on Vt are discussed in
The simple Square-Law form of Id equations (Brews, 1978) in linear region and
W ⎡ 1 2⎤
Id = μCox ⎢⎣(Vg − Vt )Vd − Vd Vd < Vdsat
2 ⎥⎦
when (2)
L
W ⎡ (Vg − Vt ) 2 ⎤
Id = μCox ⎢ ⎥ when Vd ≥ Vdsat (3)
L ⎣ 2 ⎦
where Vdsat = (Vg − Vt ) , μ is the carrier mobility, Cox is the gate oxide capacitance, W is
are bulk charge effect (Abulk), CLM, DIBL, field dependent carrier’s mobility, µ, SCBE,
Ioff is modeled using Pvoff, which is the subthreshold offset voltage parameter.
when Vg is below Vt. These effects are further discussed in the following sections.
Isub, which is caused by SCBE is model in BSIM3v3. The parameters that used
to model the Isub are actually Palpha0, Palpha1 and Pbeta0, but not Ppscbe1 and Ppscbe2
5
(although these last two parameters’ names have the term SCBE). SCBE is actually
caused by impact ionization that produces extra electrons that sweep into the drain that
increases Id and also produces extra holes that sweep into the substrate that increases
Isub. The increase of Id is modeled using Ppscbe1 and Ppscbe2 and the increase of Isub is
The body effect refers to the effect of Vb on Vt. The body effect increases the Vt
the threshold voltage because the channel depletion layer becomes wider and hence
where Pk1 is the first-order body effect parameter and Pk2 is the second-order body
effect parameter. Default values of Pk1 = 0.53 V1/2 and Pk2 = -0.0186. From Equation 4,
the more positive value of Pk1 and Pk2, the more significant the body effect is. φ f is the
difference between Fermi potential and the intrinsic potential of the substrate. 2φ f of a
The bulk charge effect is closely related to the body bias effect and refers to the
changing Vt along the channel when Vd > 0 V. Vt is not constant along the channel
because the width of the depletion region along the channel is not uniform in the
6
Figure 1.1: Non-uniform depletion width when Vd > 0 V (Cheng and Hu, 1999).
⎡ ⎡ Pa0 Leff ⎤⎤
⎢ ⎢ ⎥⎥
⎢ ⎢ Leff + 2 Pxj Xdep ⎥⎥
⎢ ⎢⎛ 2 ⎥⎥
⎢ ⎢⎜ ⎛ ⎞ ⎞⎟⎥ ⎥
Pk1 ⎜ ⎟ ⎥
⎢⎜1 − Pags (Vg − Vt )⎜
Leff 1
Abulk = ⎢1 + ⎟ ⎥ (5)
⎢ 2 2φ f − Vb ⎟ 1 + PketaVb
⎢⎜
⎝ ⎝ Leff + 2 Pxj Xdep ⎠ ⎟⎠⎥ ⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ Pb0 ⎥⎥
⎢ ⎢+ Weff + P ⎥⎥
⎣⎢ ⎢⎣ b1
⎦⎥ ⎦⎥
where Pa0 is the L dependence bulk charge parameter, Pags is gate bias dependence
bulk charge parameter, Pketa is body bias dependence bulk charge parameter, Pb0 is
channel width bulk charge parameter, Pb1 is channel width offset bulk charge
parameter. The default values of Pa0 = 1, Pags = 0, Pketa = -0.047 V-1, Pb0 = 0 m, Pb1 = 0
m. Pxj is source/drain junction depth parameter and Xdep is the channel depletion
thickness in the substrate. Weff is the effective channel width. The Weff is further
Equation 6 and Equation 7 show the Id equation with the term Abulk.
W ⎡ 1 ⎤
Id = μCox ⎢⎣ (V g − Vt )Vd − Abulk ⋅ Vd 2 ⎥ when Vd < Vdsat (6)
L 2 ⎦
7
W ⎡ (V g − V t ) 2 ⎤
I d = μ C ox ⎢ ⎥ when Vd ≥ Vdsat (7)
L ⎣ 2 A bulk ⎦
(Vg − Vt )
where Vdsat =
Abulk
SCE is especially pronounced when Vd = Vdd (or at high drain bias). Figure 1.2(a)
shows the long channel depletion region, where the effect from the source and drain
can be ignored.
depletion region to the channel. This results in electric field pattern in two-dimensional,
as shown in Figure 1.2(b). In other words, the depletion charge under the gate is
actually induced by the gate together with the source and drain. The channel charge is
considered to be “shared” by the gate, source and drain (Charge Sharing Model).
Hence, less gate charge density (smaller gate voltage) is needed to induce inversion in
short channel transistors than in long channel transistors. This means the gate voltage,
Figure 1.2: Charge regions: (a) in long channel; (b) in short channel (Liu, 2001).
8
SCE effect on Vt is model in BSIM3v3 by using the Equation 8,
⎢ ⎥ (8)
⎣ ⎦
ε s Xdep
where Lt = (1 + Pdvt2Vb ) and
Cox
Pdvt0 is the first SCE parameter, Pdvt1 is the L dependence exponent SCE parameter
and Pdvt2 is the body bias SCE parameter. The default value of Pdvt0 = 2.2, Pdvt1 = 0.53
and Pdvt2 = -0.032 V-1. Leff is the effective channel length. The Leff is further discussed in
Section 1.3.17. Vbi is the built-in voltage of the source-bulk/drain-bulk junction and εs is
uniform lateral doping effect, due to pocket implantation, as shown in Figure 1.3. The
channel doping concentration near the source and drain is higher than in the middle of
the channel. The increased doping concentration in the two ends of the channel can
result in an increase in Vt as the channel becomes shorter. The combined RSCE and
1.4.
⎛ P ⎞
ΔVt , rsce = Pk1 ⎜⎜ 1 + nlx − 1⎟⎟ 2φ f (9)
⎝ Leff ⎠
9
where Pnlx is the RSCE parameter and Leff is the effective channel length. The default
value of Pnlx = 1.74 x 10-7 m. RSCE increases the Vt of short channel transistors and
Figure 1.3: Lateral doping profile with pocket implantation (Cheng and Hu, 1999).
Figure 1.4: Vt versus L characteristics of a transistor with RSCE (Cheng and Hu, 1999).
10
1.3.4 Narrow Width Effect (NWE)
technology is used. Figure 1.5 shows the transistor cross-section in the channel width,
W direction with LOCOS isolation. Both ends of the gate oxide (in W direction) are
known as “bird beak” due to their shape. The “bird beak” shape causes thicker gate
oxide at the edge and hence results in higher Vt. As W reduces the “bird beak” effect
Figure 1.5: CMOS transistor with LOCOS isolation technology (Liu, 2001).
Also the fringing field from the gate that terminates at bulk charges outside the
intrinsic portion of the transistor (as defined by Weff), as shown in Figure 1.5, also
causes higher Vt. This is because it takes a larger gate voltage to deplete the extra bulk
where Pk3 is the NWE parameter, Pk3b is the body effect NWE parameter, Ptox is the
oxide thickness parameter, Pw0 is the channel width offset NWE parameter and Weff is
11
the effective channel width. The default value of Pk3 = 80, Pk3b = 0 and Pw0 = 2.5 x 10-6
m.
NWE only happens when LOCOS isolation is used. From quarter micron
technology and below, the LOCOS isolation has been replaced by STI. STI process
causes the opposite effect on Vt, which is known as INWE. Hence, Pk3 should be set to
a smaller value, such as 0.001 to make this term negligible in the final Vt calculation.
Unfortunately, the INWE that decreases Vt for narrow width transistors is not captured
technology is used. The STI profile causes thinner gate oxide at the edge of the
channel and hence smaller Vt, as shown in Figure 1.6. When W reduces, the thinner
gate oxide at the edge of the channel causes the entire channel to have lower Vt.
The field lines from the gate electrode that are focused by the sharp geometry
of the channel edge also lower the Vt. This is because at the edges of the channel, an
inversion layer is formed at a lower voltage than at the center. As a result, the average
previous section, INWE is not modeled in both BSIM3v3 and BSIM4. Therefore,
12
1.3.6 Small Size Effect (SS) on Threshold Voltage
the transistor Vt as L is reduced from SCE, the SS correct the short channel Vt for
⎢ ⎥ (11)
⎣ ⎦
ε s X dep
where Ltw = (1 + Pdvt2wVb ) and
Cox
Pdvt0w is the first SS parameter, Pdvt1w is the L dependence exponent SS parameter and
Pdvt2w is the body bias SS parameter. The default value of Pdvt0w = 0, Pdvt1w = 5.3e6 m-1
The potential barrier at the surface between the source and drain for NMOS
transistor is shown in Figure 1.7. Under off conditions, the potential barrier in the p-type
region prevents electron from flowing to the drain. The surface potential is mainly
When the Vg is below Vt, there are only a limited number of electrons injected
from the source over the barrier and flow into the drain. This is called subthreshold
current. In the long channel case, the potential barrier is flat over most of the transistor
because the source and drain fields only affect the very ends of the channel.
As L become shorter, the source and drain fields penetrate deeply into the
middle of the channel, which lowers the potential barrier between source and drain.
This causes a substantial increase of the subthreshold current. In other words, the Vt
13
becomes lower than that of the long channel value. The region of maximum potential
barrier also shrinks to a single point near the center of the transistor.
lowered even more, resulting in further Vt decrease. The maximum point of the barrier
also shifts toward the source end as shown in Figure 1.7. This effect is referred to as
The conduction band profile of the DIBL effect on short channel transistor is
shown in Figure 1.8. The electrons in the source region, which is at lower potential flow
to the drain region (at higher potential) after overcoming the gate controlled barrier.
Large Vd lowers the barrier and thus less gate voltage is needed to provide the
electrons enough energy to overcome the barrier and flow to the drain region.
Figure 1.7: Surface potential versus L for (a) long channel transistor (b) short channel
transistor, and (c) short channel transistor at high Vd (Taur and Ning, 1998).
14
Figure 1.8: Conduction band profile showing DIBL effect in short channel transistor
(Yeo, 2005).
ε s Xdep,0
where Lt0 = and
Cox
Pdsub is the Vt L dependence exponent DIBL parameter, Peta0 is the Vt DIBL parameter
and Petab is the Vt body bias DIBL parameter. The default value of Pdsub = 0.56, Peta0 =
0.08 and Petab = -0.07 V-1. Xdep,0 is the depletion thickness in the substrate at zero bulk
bias.
15
1.3.8 Channel Length Modulation (CLM)
The saturation of Id, Idsat can be understood from the inversion charge density.
When Vd is small (linear region), the inversion charge density at the drain end of the
channel is only slightly lower than the source end, as shown in Figure 1.9(a).
As the Vd increases (for a fixed Vg), the Id increases, but the inversion charge
density at the drain decreases until finally it goes to zero when Vd = Vdsat = Vg – Vt. At
this voltage, Id reaches its maximum value. In other words, the surface channel
vanishes at the drain end of the channel when saturation occurs. This is called pinch-
When Vd increases beyond the saturation, the pinch-off point moves toward the
source, but the Id remains essentially the same. This is because for Vd > Vdsat, the
voltage at the pinch-off point remains at Vdsat and the current stays the same apart from
a slight decrease in L (to L’), as shown in Figure 1.9(c). This phenomenon is called
Channel Length Modulation (CLM) that makes the channel look as if it is shorter.
Since Id increases when L reduces, the CLM increases the Idsat. This effect can
be seen from the plots of Id-Vd curves. CLM and DIBL effect on Id is modeled in
⎛ Vd − Vdsat ⎞
Id = Id,0⎜1 + ⎟ (13)
⎝ Va ⎠
variable that adjusts the slope of the variation of Id with respect to Vd. Graphically, can
be taken to be value on the –x axis for which the current extrapolates to 0 A, as shown
in Figure 1.10.
16
Figure 1.9: CMOS transistor operates in; (a) linear region, (b) onset of saturation,
(c) beyond saturation where CLM occurs (Taur and Ning, 1998).
17
Figure 1.10: Id-Vd curves with Vg,2 > Vg,1, showing the relationship between Va and the
Va,sat and Va,clm_dibl equations are shown in the following Equation 14 and
Equation 15:
μeff
1 + Ppvag (Vg − Vt )
2 Pvsat Leff
Va , clm _ dibl =
Ppclm f1 + θrout (1 + PpdiblcbVb ) f 2 (15)
f0, f1 and f2 represent complicated functions and other variables that are covered in this
velocity parameter. Ppclm is channel length modulation parameter, Ppdiblc1 is first DIBL
Ppdiblcb is the body bias DIBL correction on Va parameter. The default value of Ppvag = 0,
Pvsat = 8 x104 ms-1, Ppclm = 1.3, Ppdiblc1 = 0.39, Ppdiblc2 = 0.0086 and Ppdiblcb = 0.39.
18
1.3.9 Velocity Saturation
explained by the concept of channel pinch-off. For short channel transistors, the Idsat is
When the Vd increases in a long channel transistor, the Id first increases, and
then becomes saturated at a voltage with the onset of pinch-off at the drain. In a short
channel transistor, the Idsat occurs at a much lower voltage due to velocity saturation, as
shown in the Id-Vd curves in Figure 1.11. This is because at high (horizontal) field
strength, the velocity of the carriers tends to saturate due to scattering effects
(collisions suffered by the carriers). The transistor W/L is 9.5 μm/0.25 μm.
This parameter when used in NMOS, denotes the electron saturation velocity and
when used in PMOS, denotes the hole saturation velocity. Therefore, Pvsat in NMOS
should be slightly larger than the Pvsat in PMOS. Pvsat is a critical parameter determining
the transistor current in short channel transistors, but has negligible impact on long
channel transistors.
This agrees with the physical effects that in long channel transistors, the electric
field parallel to the current conduction is small and the electron drift velocity is roughly
equal to the mobility times the field. The drift velocity never reaches a magnitude
comparable to Pvsat. In short channel transistors, in contrast, the field is large enough
such that the carriers travel at the saturation velocity in a sizable portion of the channel.
Pvsat also affects Va, as shown in Equation 15 (in the previous section).
19
Figure 1.11: Id-Vd characteristics with velocity saturation, solid lines and as if there were
transistor state from “on” to “off”. In an ideal case, the subthreshold slope should be
When the subthreshold slope is degraded, this usually means that the transistor
is harder to turn “off” because the gate loses control over the channel surface potential.
The subthreshold slope degradation also reduces the Vt and causes higher Ioff.
20
Figure 1.12: Subthreshold characteristics of short and long channel transistors at low
Subthreshold conduction effect on Id at off condition (Vg < Vt), Ioff is modeled in
2 ⎛ qVt ⎞ ⎛ qPvoff ⎞
⎜− ⎟
W ⎛ kT ⎞ ⎜⎜ −
⎝ ηkT
⎟⎟ ⎜ ηkT ⎟
Ioff = μCox ⎜ ⎟ e ⎠
e ⎝ ⎠
L ⎜⎝ q ⎟⎠ (16)
where Pvoff is the subthreshold offset voltage parameter. The default value of Pvoff =
-0.08 V. q is the single electron charge (1.6e-19 C). From Equation 16, Ioff is
⎛ qV g ⎞
⎜⎜ ⎟⎟
⎝ η kT
Ioff ∝ e ⎠
(17)
21
The ideality factor, η is given in Equation 18,
⎛ Cdep ⎞ Pcit
η = 1 + ⎜ Pnfactor ⎟+ +
⎝ Cox ⎠ Cox
⎡ ⎛ ⎛ L ⎞
⎜⎜ − Pdvt1 eff ⎟⎟ ⎞ ⎤
⎛ L ⎞
ε s Xdep
where Lt = (1 + Pdvt2Vb ) and
Cox
Pnfactor is subthreshold turn-on swing parameter, Pcit is the interface trap capacitance
parameter (this parameter is purely a DC parameter which does not affect the C-V
pure DC parameter). The default value of Pnfactor = 1, Pcit = 0 Fm-2, Pcdsc = 2.4e-4 Fm-2,
Pcdscd = 0 FV-1m-2 and Pcdscb = 0 FV-1m-2. Cdep is the depletion capacitance per unit area
in the bulk.
The last term is similar to the one used in SCE for Vt calculation. It is meant for
the short channel transistors. Usually to model the subthreshold current, Pnfactor and
Pvoff are used. Other parameters can set to their default values, rule of thumb is the
takes place along the interface between silicon, Si and silicon dioxide, SiO2. The carrier
mobility at the interface is lower than in the bulk and depends on both vertical and
22
In the early years of CMOS, the gate oxide was thick, and the vertical electric
field induced by the gate bias was low and hence the influence of the vertical electric
field could be ignored. But in today’s short channel transistors, oxide thicknesses are
very thin, for example in 0.13 μm technology the oxide thickness is about 2 nm. Thus,
the influence of the vertical electric field is strong and the carrier mobility is not constant
dependence of mobility on the vertical component of the electric field. They are phonon
temperature and bias. For good quality interfaces, phonon scattering is generally the
Figure 1.13: Mobility versus Vertical Effective Field to illustrate how the mobility
23
In BSIM3v3, the effective carrier mobility, μeff, of mobility mode parameter,
Pu0
μeff = 2 (19)
⎛ Vg + Vt ⎞ ⎛ Vg + Vt ⎞
1 + (Pua + PucVb )⎜⎜ ⎟⎟ + Pub ⎜⎜ ⎟⎟
⎝ tox ⎠
P ⎝ tox ⎠
P
where Pu0 is the zero field universal mobility parameter, Pua is the first order mobility
degradation parameter, Pub is the parabolic mobility degradation parameter and Puc is
the body bias mobility degradation parameter. The default value of Pu0 = 0.067 m2V-1s-1
(NMOS) and 0.025 m2V-1s-1 (PMOS), Pua = 2.25 x 10-9 mV-1, Pub = 5.87 x 10-19 m2V-2
At saturation mode, the high electric field near the drain region causes impact
ionization of carriers. For NMOS transistor, the generated electrons are swept into the
drain whereas the holes flow into the substrate. This phenomenon is known as SCBE.
SCBE results in Id increase that is many times larger than Isub, as shown in Figure 1.14.
Figure 1.14: Id-Vd characteristics due to SCBE (Cheng and Hu, 1999).
24